One hundred ninety-four patients with cryptococcal meningitis were enrolled in a multicenter, prospective, randomized clinical trial to compare the efficacy and toxicity of four as compared with six weeks of combination amphotericin B and flucytosine therapy. Among 91 patients who met preestablished criteria for randomization, cure or improvement was noted in 75 percent of those treated for four weeks and in 85 percent of those treated for six weeks. The estimated relapse rate for the four-week regimen was higher--27 as compared with 16 percent--whereas the incidence of toxic effects for the two regimens was similar--44 as compared with 43 percent. Among 23 transplant recipients, 4 of 5 treated for four weeks relapsed, leading to the decision to treat the rest of the group for six weeks. Only 3 of the 18 treated for six weeks relapsed. In a third group of 80 patients, the protocol was not followed during the initial four weeks, and these patients were not randomized. Thirty-eight died or relapsed. Multifactorial analysis of pretreatment factors for all 194 patients identified three significant predictors (P less than 0.05) of a favorable response: headache as a symptom, normal mental status, and a cerebrospinal fluid white-cell count above 20 per cubic millimeter. These and other findings in this study are consistent with the view that the four-week regimen should be reserved for patients who have meningitis without neurologic complications, underlying disease, or immunosuppressive therapy; a pretreatment cerebrospinal fluid white-cell count above 20 per cubic millimeter and a serum cryptococcal antigen titer below 1:32; and at four weeks of therapy, a negative cerebrospinal fluid India ink preparation and serum and cerebrospinal fluid cryptococcal-antigen titers below 1:8. Patients who do not meet these criteria should receive at least six weeks of therapy.